Econometrics and Lomb-Scargle

jslade · Post by **jslade** » Sat Oct 19, 2019 6:24 pm

Does anyone have a better periodicity finder than Lomb-Scargle? It's always worked for me, but I feel like with the advent of wavelets and sparse/compressive sensing techniques, there should be something better by now.

This leads to the larger question; are there any references out there for "econometric" or signal processing techniques which incorporate proven "newer" ideas? I mean, the classics work, but I'm assuming there is newish stuff worth knowing, which can be found in handy book-like references rather than collections of papers. Generally speaking I pick up Stephen Taylor's ancient book when I want to think along these lines ... seems weird using a book written in 1986.

nikol · Post by **nikol** » Sat Oct 19, 2019 7:27 pm

google: signal processing ai

for example

https://medium.com/@alexrachnog/deep-learning-the-final-frontier-for-signal-processing-and-time-series-analysis-734307167ad6

Jurassic · Post by **Jurassic** » Sun Oct 20, 2019 11:21 am

@nikol thanks for sharing, the article was a good read

Maggette · Post by **Maggette** » Mon Oct 21, 2019 8:24 am

Not sure if a wild medium blog posts with a lot of ANN buzzword bingo is what jslade was looking for;)...though I have to say that I share similiar experiences wregarding the suprisingly good performance of CNNs.

As a note: the most I have learned (and I am still learning) about deep learning in time series data was while using data that I generated myself. It is extremly instructive if you know the underlying data generating proccess....and than see what can go wrong:)

@jslade: recently I played around with Empirical Mode Decomposition (EMD) a lot lately and have something in production running with it. Running EMD and fitting something stupid on EMDs worked fine for me.

nikol · Post by **nikol** » Mon Oct 21, 2019 6:22 pm

What is a "classic" method? It is a "wildly" acceptable method.

Nah. Giving medium mediocrity grade is pride and vanity. I guess one need just a "newer" idea, while the rest can be elaborated by skilled professional and being made as complex as he/she pleases. People in medium share nice ideas, that's what it is for.

jslade · Post by **jslade** » Fri Oct 25, 2019 10:14 pm

That article was shit Nikol; not even in the ballpark of econometric technique, and a poor exposition of neural doohickeys to boot. Piss

Thanks M for reminding me about EMD/Hilberty things; I've been meaning to screw around with them again, and indeed might be a good thing to use in place of Lomb periodograms.

Really: there must be a classic econometrics or signal processing book written in the 21st century. Maybe everyone too busy fiddling with Yann's white elephants to notice.

nikol · Post by **nikol** » Sat Oct 26, 2019 3:43 pm

> That article was shit Nikol

@jslade

EDIT: removed bravado response. stupid display of my own pride and vanity [go to confession corner]